EP2352148B1

EP2352148B1 - Noise gate, sound collection device, and noise removal method

Info

Publication number: EP2352148B1
Application number: EP09827585.2A
Authority: EP
Inventors: Ryo Tanaka; Naoto Kuriyama
Original assignee: Yamaha Corp
Current assignee: Yamaha Corp
Priority date: 2008-11-21
Filing date: 2009-11-18
Publication date: 2013-10-16
Anticipated expiration: 2029-11-18
Also published as: EP2352148A4; CN102216984A; KR101254876B1; KR20110082176A; EP2352148A1; US9036830B2; CN102216984B; US20110268288A1; JP2010122617A; WO2010058804A1

Description

Technical Field

The present invention relates to a noise gate for removing a noise included in an audio signal, a sound collection device and a noise removing method.

Background Art

A sound collection device of the related art is provided with a noise gate as a unit for removing a noise (e.g., JP-A-2006 50067 ).
For example, JP-A-2006 50067 discloses a noise gate device that changes a threshold value of a noise gate in accordance with an amplification degree of a head amplifier for amplifying an audio signal and removes an audio signal less than the threshold value.
US 4,811,404 A discloses a noise suppression system which performs speech quality enhancement upon the speech-plus-noise signal available at the input to generate a clean speech signal at the output by spectral gain modification. The measures taken include the addition of a signal-to-noise ratio (SNR) threshold mechanism to reduce background noise flutter by offsetting the gain rise of the gain tables until a certain SNR threshold is reached, the use of a voice metric calculator to produce more accurate background noise estimates via performing the update decision based on the overall voice-like characteristics in the channels and the time interval since the last update, and the use of a channel SNR modifier to provide immunity to narrowband noise bursts through modification of the SNR estimates based on the voice metric calculation and the channel energies.
WO 87/00366 A1 discloses a noise suppression system which performs speech quality enhancement upon speech-plus-noise signal available at the input to generate a clean speech signal at the output by spectral gain modification. The noise suppression system includes a background noise estimator which generates and stores an estimate of the background noise power spectral density based upon pre-processed speech, as determined by the detected minima of the post-processed speech energy level. This post-processed speech may be obtained directly from the output of the noise suppression system, or may be simulated by multiplying the pre-processed speech energy by the channel gain values of the modification signal. The channel gain controller produces these individual channel gain values for application to both the channel gain modifier and the background noise estimator. Each individual channel gain value is selected as a function of (a) the channel number, (b) the current channel SNR estimate, and (c) the overall average background noise level. The technique of implementing post-processed signal to generate the background noise estimate provides a more accurate measurement of the background noise energy since it is based upon much cleaner speech signal. As a result, the present invention performs acoustic noise suppression in high ambient noise backgrounds with significantly less voice quality degradation.

Summary of Invention

Problems that the Invention is to Solve

However, the noise gate device described in JP-A-2006 50067 removes an audio signal less than a threshold value. Therefore, the problem arises that in a case where an utterance voice of a speaking person is included in the audio signal less than the threshold value, the noise gate device removes the utterance voice having a signal level less than the threshold value.
In view of the above problem, a purpose of the invention is to provide a noise gate, a sound collection device and a noise removing method each capable of readily removing a stationary noise without degrading an audio signal.

Means for solving the Problems

A noise gate according to the invention is provided as set forth in claim 1 and estimates a noise spectrum of a stationary noise included in an input audio signal (hereinafter, referred to as the stationary noise) based on a frequency spectrum of the input audio signal. The noise gate computes a signal level ratio of the frequency spectrum of the input audio signal to the estimated noise spectrum. The noise gate outputs the frequency spectrum of the audio signal in which the signal level in a frequency band is decreased by a certain level in the frequency band having the computed signal level ratio which is less than a predetermined threshold value and outputs said spectrum having the signal level maintained in the frequency band having the signal level ratio equal or greater than said threshold.
With this configuration, the noise gate outputs a frequency spectrum of an audio signal including an utterance voice of a speaking person and suppresses a signal level in a frequency spectrum of an audio signal formed of a stationary noise so that the audio signal in which only the stationary noise is removed can be output without degrading the utterance voice of the speaking person.
Further, the threshold value can be set to a value greater than 1 which is a value compared to the signal level ratio by a value corresponding to an estimation error of an estimation unit. Since the noise gate computes the signal level ratio to be compared to the threshold value by using the estimated stationary noise, the estimation error of the stationary noise is considered.
With the above configuration, the noise gate can remove the stationary noise even in a case where the signal level of the stationary noise is estimated to be a value less than an actual value.
Moreover, a sound collection device according to the invention is provided as set forth in claim 2 and has the above described noise gate. The sound collection device transforms an audio signal generated by collecting a sound into a frequency spectrum as a signal of a frequency domain and outputs it to the noise gate. In addition, the sound collection device inversely transforms a frequency spectrum of an audio signal output from the noise gate into an audio signal as a signal of a time domain.
With the above configuration, the sound collection device can remove a stationary noise from the collected audio signal by using the above described noise gate. Therefore, the sound collection device can output a sound without degrading a collected utterance voice of a speaking person.
A noise removing method according to the invention is provided as set forth in claim 3 and includes a process of inputting a frequency spectrum of an audio signal, a process of estimating a noise spectrum based on the frequency spectrum of the audio signal input in the inputting process, a process of computing a signal level ratio of the frequency spectrum of the audio signal to the noise spectrum estimated in the estimating process, and a process of outputting the frequency spectrum of the audio signal in which the gain in a frequency band having the signal level ratio less than a predetermined threshold value is decreased by a certain level, and outputs said spectrum having the signal level maintained in the frequency band having the signal level ratio equal or greater than said threshold, the signal level ratio being computed in the computing process.
In addition, the threshold value can be set to a value greater than 1 which is a value compared to the signal level ratio by a value corresponding to an estimation error of an estimation unit used in the estimating process.

Advantage of the Invention

The noise gate according to the invention can output an audio signal in which only a stationary noise is removed without degrading an utterance voice of a speaking person.

Brief Description of the Drawings

[Fig. 1] Fig. 1 is a block diagram showing a function and a structure of a sound collection device.
[Fig. 2] Fig. 2 is a graph showing an example of a gain table.
[Fig. 3] Figs. 3(A), 3(B) and 3(C) are graphs respectively showing examples of a stationary noise, an audio signal and an audio signal after removing a noise.
[Fig. 4] Fig. 4 is a graph showing another example of a gain table.

Mode for Carrying Out the Invention

A sound collection device 1 according to the invention is described below with reference to Figs. 1 and 2. Fig. 1 is a block diagram showing a function and a structure of the sound collection device 1. Fig. 2 is a graph showing an example of a gain table. The sound collection device 1 has a noise gate 12 (see Fig. 1). The sound collection device 1 transforms a collected audio signal NE'T into a frequency spectrum NE'N. The sound collection device 1 removes a stationary noise N'N included in the audio signal NE'N by using the noise gate 12. At that time, the stationary noise N'N is estimated based on the frequency spectrum NE'N by the noise gate 12. The sound collection device 1 outputs an audio signal CO'T which is generated in such a manner that a frequency spectrum CO'N after removing the stationary noise N'N is inversely transformed with respect to a time axis. Meanwhile, an end symbol of a signal in a time domain is indicated by 'T and an end symbol of a signal in a frequency domain is indicated by 'N in the descriptions below.
First, the function and the structure of the sound collection device 1 are described with reference to Figs. 1 and 2. As shown in Fig. 1, the sound collection device 1 includes a microphone MIC, an FFT processing unit 11, the noise gate 12 and an IFFT processing unit 13.
The microphone MIC generates an audio signal by collecting ambient sounds. The microphone MIC outputs the generated audio signal NE'T to the FFT processing unit 11.
The FFT processing unit 11 is a fast Fourier transform circuit that transforms the audio signal NE'T as the signal in the time domain into the audio signal NE'N as the signal in the frequency domain and outputs the audio signal NE'N to an estimation unit 121 and a noise removal unit 122 in the noise gate 12.
The noise gate 12 has the estimation unit 121 and the noise removal unit 122.
The estimation unit 121 performs a process of estimating a stationary noise N'N included in the audio signal NE'N input from the FFT processing unit 11. The estimation unit 121 sequentially acquires frequency spectrums (hreinafter, referred to as the audio spectrum(s)) S (NE'N) at certain sampling times of the stationary noise N'N and stores them temporarily. The estimation unit 121 estimates frequency spectrums (hereinafter, referred to as the noise spectrum(s)) S (N'N) at certain sampling timings of the stationary noise N'N based on the plurality of times of audio spectrums S (NE'N) acquired and stored as in the above. The estimation unit 121 outputs the estimated noise spectrums S(N'N) to the noise removal unit 122.
For example, the noise spectrum at a certain sampling timing T is represented by S(N'N(T)), the audio spectrum at the same sampling timing T is represented by S(NE'N(T)), and the noise spectrum at the previous sampling timing T-1 is represented by S(N'N(T-1)). "α" and "β" are forgetting constants and they are, for example, set to be α=0.9 and β=0.1. The noise spectrum S(N'N(T)) is represented by the following formula 1.
$S (NʹN (T)) = αS (NʹN (T - 1)) + βS (NEʹN (T))$

Thus, by estimating the noise spectrum S(N'N(T)) based on the audio spectrum, a stationary noise such as a background noise or the like can be estimated. The estimation unit 121 performs an estimating process of a noise spectrum only in a case where a level of an audio signal collected by the microphone MIC is in a low state (a silent state).
The noise removal unit 122 removes the stationary noise N'N from the audio signal NE'N input from the FFT processing unit 11 and outputs the audio signal CO'N after removing the noise to the IFFT processing unit 13. To be specific, the noise removal unit 122 computes a signal level ratio of the audio spectrum S(NE'N) to the noise spectrum S(N'N) input from the estimation unit 121. In a case where the computed signal level ratio is less than the threshold value, the noise removal unit 122 decreases the gain of the audio spectrum S(NE'N) in a frequency band having the signal level ratio less than the threshold value, and outputs the audio signal.
For example, the noise removal unit 122 has a gain table TBL shown in Fig. 2. The gain table TBL determines the gain of the audio spectrum S(NE'N) in accordance with the computed signal level ratio. The gain table TBL makes the gain of the audio spectrum S(NE'N) to be 1 in a case where the signal level ratio is equal to or greater than the threshold value and makes the gain of the audio spectrum S(NE'N) to be 0 in a case where the signal level ratio is less than the threshold value. In this case, the frequency spectrum S(CO'N) at a certain sampling timing of the audio signal CO'N after removing the noise can be represented by the following formula 2.
$S (COʹN) = (TBL (S (NEʹN) / S (NʹN)) \cdot S (NEʹN))$

As a result, in a frequency band in which the signal level ratio is equal to or greater than the threshold value, the noise removal unit 122 outputs the audio spectrum S(NE'N) as it is without changing the signal level of the audio spectrum S(NE'N). Alternatively, in a frequency band in which the signal level ratio is less than the threshold value, the noise removal unit 122 outputs the audio spectrum S(NE'N) having a signal level which is zero since the audio spectrum S(NE'N) is assumed as the noise spectrum S(N'N).
In addition, by considering an estimation error of the stationary noise N'N, the noise removal unit 122 may set a threshold value of the signal level ratio to a value being slightly greater than the threshold value (1.0) of the signal level ratio. With this, the noise removal unit 122 can remove the stationary noise N'N even in a case where the signal level of the stationary noise N'N is estimated to be less than an actual level because of the estimation error of the stationary noise N'N.
The IFFT processing unit 13 is an inverse fast Fourier transform circuit that transforms the audio signal CO'N as a signal of a frequency domain into an audio signal CO'T as a signal of a time domain.
With the above, since the sound collection device 1 can leave the audio signal as it is without attenuating the audio signal in the frequency band having an utterance voice of a speaking person, the utterance voice of the speaking person can be output without being degraded.
The audio signal CO'N after removing the noise as the output value of the noise gate 12 is described with reference to Figs. 3(A) to 3(C). Figs. 3(A) to 3(C) are graphs respectively showing examples of the stationary noise, the audio signal and the audio signal after removing the stationary noise. Fig. 3(A) indicates a signal level of the stationary noise, Fig. 3(B) indicates a signal level of the audio signal including the stationary noise, and Fig. 3(C) indicates a signal level of the audio signal after removing the stationary noise.
The sound collection device 1 estimates the stationary noise N'N as shown in Fig. 3(A) and generates the audio signal NE'N as shown in Fig. 3(B). When the sound collection device 1 simply subtracts the noise N'N from the audio signal NE'N in order to remove the stationary noise N'N from the audio signal NE'N, the signal level of the audio signal NE'N is attenuated even in a frequency band having an utterance voice of a speaking person as shown by a dotted line in Fig. 3(C) so that the sound collection device 1 may output a sound having degraded quality. The sound collection device 1 may have a case where the stationary noise N'N remains in the audio signal NE'N because of the estimation error of the stationary noise N'N. In this case, the sound collection device 1 may generate a musical noise because the signal level is remarkably varied in a frequency band not having an utterance voice of a speaking person.
In the embodiment, since the sound collection device 1 leaves a frequency band having an utterance voice of a speaking person as shown by a solid line in Fig. 3(C) and decreases the signal level (the gain) of the audio signal in a frequency band not having an utterance voice of a speaking person (a frequency band only having a noise), the sound collection device 1 can output a sound without degrading the quality of the utterance voice of the speaking person. In addition, since the sound collection device 1 determines the output value CO'N in accordance with the signal level ratio of the audio signal NE'N to the stationary noise N'N, variation in the signal level is not liable to occur in the frequency band not having an utterance voice of a speaking person and occurrence of the musical noise can be prevented. Further, since the sound collection device 1 computes the signal level ratio of the audio signal NE'N to the stationary noise N'N, a frequency band only having a noise can be detected even when a level of an utterance voice of a speaking person is low.
In the above embodiment, the sound collection device 1 maintains the signal level of the audio signal in the frequency band having an utterance voice of a speaking person and makes the signal level (the gain) of the audio signal in the frequency band not having an utterance voice of a speaking person to be 0, and then outputs the audio signal. However, the sound collection device 1 can maintain the signal level of the audio signal in the frequency band having an utterance voice of a speaking person and can decrease the signal level (the gain) of the audio signal in the frequency band not having an utterance voice of a speaking person to be a value greater than 0. Fig. 4 is a graph showing another example of a gain table. In this case, for example, as shown in Fig. 4, the sound collection device 1 determines the gain of the audio signal in the frequency band not having an utterance voice of a speaking person to be 0.5 and outputs the audio signal in the frequency band not having an utterance voice of a speaking person by multiplying 0.5 to the signal level of the audio signal. With this, since the stationary noise in the frequency band having an utterance voice of a speaking person is not remarkable, the sound collection device 1 can naturally output the voice of the speaking person.
While the invention is described in detail by referring to the specific embodiment, it is understood by those of ordinary skill in the art that various modifications and changes can be made without departing from the scope of the invention as defined by the appended claims.

Industrial Applicability

It is possible to provide a noise gate that outputs an audio signal in which only a stationary noise is removed without degrading an utterance voice of a speaking person.

Claims

A noise gate (12) comprising:
an input unit to which a frequency spectrum of an audio signal is input;

an estimation unit (121) adapted to estimate a noise spectrum based on the frequency spectrum of the audio signal input to the input unit; and characterized by further comprising:
an output unit adapted to compute a signal level ratio of the frequency spectrum of the audio signal to the noise spectrum estimated by the estimation unit (121) and to output the frequency spectrum of the audio signal in which the signal level in a frequency band having the signal level ratio less than a predetermined threshold value is decreased by a certain level and the signal level in the frequency band having the signal level ratio equal to or greater than the predetermined threshold is maintained.
A sound collection device (1) comprising:
the noise gate (12) according to claim 1;

a sound collection unit (MIC)adapted to collect ambient sounds and to generate an audio signal;

a transformation unit (11) adapted to transform the audio signal generated by the sound collection unit (MIC) into a frequency spectrum as a signal of a frequency domain and arranged such that the frequency spectrum transformed by the transformation unit (11) is input into the input unit of the noise gate (12); and

an inverse transformation unit (13) adapted to inversely transform the frequency spectrum of the audio signal output by the output unit of the noise gate (12) into an audio signal as a signal in a time domain.
A noise removing method, comprising the steps of:
inputting a frequency spectrum of an audio signal;

estimating a noise spectrum based on the frequency spectrum of the audio signal input in the inputting step;

computing a signal level ratio of the frequency spectrum of the audio signal to the noise spectrum estimated in the estimating step; and characterized by further comprising:
outputting the frequency spectrum of the audio signal in which the signal level in a frequency band having the signal level ratio less than a predetermined threshold value is decreased by a certain level and the signal level in the frequency band having the signal level ratio equal to or greater than the predetermined threshold is maintained, the signal level ratio being computed in the computing step.